Biomedical Engineering Reference
In-Depth Information
Table 2.2
Model-based prediction algorithms of respiratory motion
Methods
Prediction error and evaluation metrics
Features (system)
Linear predictor [
75
]
Around 2.2 mm with 200 ms latency, RMSE
RMSE at 10 Hz (RTRT)
Kalman filter [
75
]
Around 2.5 mm with 200 ms latency, RMSE
RMSE at 10 Hz (RTRT)
Sinusoidal model [
74
]
Less than 2 mm with 200 ms latency, standard deviation
1D prediction (RPM)
Finite state model [
78
]
Less than 1.5 mm, RMSE
Three line segments (EX-EOE-IN)
(RTRT)
Vector model based on tidal volume and
airflow [
92
]
0.28-1.17 mm
Standard deviation (digital
spirometer)
Patient-specific model using PCA [
5
]
Around 2-3 mm, standard deviation
Respiration-correlated CT (RPM)
Autoregressive moving average model
[
20
,
71
]
0.8 mm with 200 ms latency, standard deviation
Image rate: 1.25-10 Hz (RTRT,
RPM)
Deformation from orbiting views [
16
]
2.5 mm (LR), 1.7 mm (SI), standard deviation
Cone-beam CT
Local regression method [
90
]
2.5 mm
Local weighted regression, RMSE
(RPM)
Optical flow deformable algorithm [
39
]
1.9 mm
Standard deviation (Philips CT
scanner)
Finite element method [
38
]
3 mm (end expiration-end inspiration), 2 mm (end expiration-mid
respiration)
Patient-specific models (Philips CT
scanner)
Surrogate-based method [
89
]
2.2-2.4 mm (carina), 3.7-3.9 mm (diaphragm)
Standard deviation (RPM)
Diaphragm-based method [
88
]
2.1 mm
Standard deviation (RPM)
Support vector regression method [
56
]
Less than 2 mm at 1000 ms latency, RMSE
30 Hz sample frequency (CyberKnife)
Quaternion-based method [
91
]
2.5 (standard deviation)
Phantom matching error (PME)
Hidden Markov model [
73
]
1.88 ms at 200 ms latency, RMSE
Various latency: 33-1000 ms (RTRT)
Kernel density estimation-based [
55
]
1.08 mm at 160 ms, 2.01 mm at 570 ms, RMSE
Multidimensional prediction
(CyberKnife)
Local circular motion model [
72
]
Less than 0.2 (nRMSE) at 200 ms normalized RMSE
First-order EKF, 5, 10, 15, 20 Hz
(RPM)
Search WWH ::
Custom Search